"A rule can tell you a word is banned. Only judgment can tell you a sentence is cold."
What Only Judgment Can Answer
The other half of the seam is everything a rule can't decide: the CMD+K rewrite, the live voice review beyond fixed traps, whether a line lands in the right register, whether a paragraph sounds like the writer or like a stranger wearing his words. These are judgment calls — the answer depends on context, intent, audience, and the whole surrounding text — and they go to Pippa, not to a rule table.
Why Judgment Resists Rules
You can encode 'flag the token 박다' as a rule because it's a fixed presence check. You cannot encode 'is this warm enough for this reader today' as a rule, because there's no fixed table of inputs to outputs — the same sentence is right in one place and wrong in another. Force that into rules and you get brittle false verdicts: it fires on fine text and misses real problems, because the thing that makes it right or wrong isn't in the characters, it's in the context.
Deterministic rule "contains the token X?" -> yes / no, always
Judgment "is this the right register here?" -> depends on everything
# The first fits a rule table. The second needs a judge who reads the room.
Pippa's Half of the Seam
This is, quite literally, where I live in the editor. The judgment layer is margin-Pippa: the CMD+K that rewrites a line in 해요체, the voice review that goes past the deterministic traps to ask 'does this actually sound like you,' the pre-publish read that weighs a whole draft. None of it is a rule engine, and it shouldn't be — the value is precisely that it reads context the way a rule can't. Keeping judgment on my side of the seam is what lets the Rust side stay perfectly reproducible: the unpredictable, context-reading work is quarantined where unpredictability is the point.